I am trying to obtain N-best lists or lattices when a finite state grammar is given as language model. I tried to obtain NBest lists with sphinx3_decode.exe and pocketsphinx_batch.exe but could not achieve to obtain lattices or Nbest lists when mode is defined as fsg. (For example lattices are generated when language model is bigram and given with option -lm biGramName but not generated when language model is fsg given with -mode fsg -fsg fsgName)
Am I doing something wrong?
Is it possible to obtain lattices or NBest lists with pocketsphinx or sphinx3_decode when mode is fsg (sphinx2 is also suitable for me if it has the capability)?
Any suggestion will be very usefull
Thank you very much
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I downloaded the latest version of pocketsphinx and obtained lattices with it in fsg mode. However I could not obtain nbest list from the same version. Is this normal? (I quickly looked at the pocketsphinx code and could not see any write function for nbest lists although there is a write function for lattices -write_lattice();-)
If there is no way of obtaining nbest lists from pocketsphinx in fsg mode, can I get best paths from generated lattice? I know that there is an executable -sphinx3_astar- to obtain nbest lists from lattices. However we should give a tri-gram language model to this program . As I could understand it rescores the lattices according to this language model and gives the best re-scored path. However I dont want to obtain re-scored path. My aim is to obtain number of best matching hypotheses in a single recognition network. How I can achieve this?
Any suggestion will be helpfull
Thanks a lot
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I am trying to obtain N-best lists or lattices when a finite state grammar is given as language model. I tried to obtain NBest lists with sphinx3_decode.exe and pocketsphinx_batch.exe but could not achieve to obtain lattices or Nbest lists when mode is defined as fsg. (For example lattices are generated when language model is bigram and given with option -lm biGramName but not generated when language model is fsg given with -mode fsg -fsg fsgName)
Am I doing something wrong?
Is it possible to obtain lattices or NBest lists with pocketsphinx or sphinx3_decode when mode is fsg (sphinx2 is also suitable for me if it has the capability)?
Any suggestion will be very usefull
Thank you very much
Thank you for your reply!
I downloaded the latest version of pocketsphinx and obtained lattices with it in fsg mode. However I could not obtain nbest list from the same version. Is this normal? (I quickly looked at the pocketsphinx code and could not see any write function for nbest lists although there is a write function for lattices -write_lattice();-)
If there is no way of obtaining nbest lists from pocketsphinx in fsg mode, can I get best paths from generated lattice? I know that there is an executable -sphinx3_astar- to obtain nbest lists from lattices. However we should give a tri-gram language model to this program . As I could understand it rescores the lattices according to this language model and gives the best re-scored path. However I dont want to obtain re-scored path. My aim is to obtain number of best matching hypotheses in a single recognition network. How I can achieve this?
Any suggestion will be helpfull
Thanks a lot
sphinx2 & sphinx3 doesn't support lattice generation in fsg mode. Latest pocketsphinx should be able to do it.